Multivariate range #39
Conversation
…o implementation should be raised automatically by ABC inheritance
…file not being found
…enable tests passing
…is outside the range
…heck that min is less than max in init
tsod/mvdetectors.py
Outdated
|
|
||
| Parameters | ||
| ---------- | ||
| min_value : float |
There was a problem hiding this comment.
min and max values must be able to be varied for each individual time-series, e.g. for simultaneous measurement of temperature and humidity, you need different ranges. But keep the option to provide a single value applied for all timeseries.
There was a problem hiding this comment.
Thanks for having a look quickly! Sure, I can do that. The intent was to have a very simple method with simple inputs. But the idea of allowing for both options (either same min/max for all or specified indvidually) is also reasonable.
The question is of course how many options to pack into one thing versus constructing another detector with this specific purpose. I think I might actually prefer pursuing the original intent and instead adding another detector that has min/max for each time series. It would also avoid some if/else in e.g. fitting, which will keep the code clearer.
Let me know what you think though, if you still prefer combining both options in one detector we can also do that.
There was a problem hiding this comment.
I think we should follow common conventions in NumPy and Pandas e.g pandas.DataFrame.clip
The clip method accepts both scalar and array as input.
There was a problem hiding this comment.
I was able to use broadcasting to avoid if/else and have added changes that allow different ranges for different time series.
…sistent with the univariate version
Add simple range detector for multivariate data. A time point is defined to be anomalous if any of the time series is outside of min/max range.